Not too close, not too far: testing the Goldilocks principle of ‘optimal’ distance in innovation networks

Abstract

This paper analyses how the formation of collaboration networks affects firm-level innovation by applying the ‘Goldilocks principle’. The ‘Goldilocks principle’ of optimal distance in innovation networks postulates that the best firm-level innovation results are achieved when the partners involved in the network are located at the ‘right’ distance, i.e. ‘not too close and not too far’ from one another, across non-geographical proximity dimensions. This principle is tested on a survey of 542 Norwegian firms conducted in 2013, containing information about firm-level innovation activities and key innovation partners. The results of the ordinal logit regression analysis substantiate the Goldilocks principle, as the most innovative firms are found among those that collaborate with partners at medium levels of proximity for all non-geographical dimensions. The analysis also underscores the importance of the presence of a substitution–innovation mechanism, with geographical distance problems being compensated by proximity in other dimensions as a driver of innovation, while there is no support for a potential overlap–innovation mechanism.

Keywords:

• Proximities
• innovation
• networks
• collaboration
• Goldilocks principle
• Norway

JEL codes:

• O31
• O33
• D85

Acknowledgements

The research for this paper has benefited from the generous funding of the Norwegian Research Council under the programmes Demosreg (grant no. 209761) and VRI (grant no. 233737), and from the European Research Council (grant no. 269868). Earlier versions of the paper were presented at the Regional Studies Association Conference in Piacenza, the DRUID Conference in Rome, and the Global Conference in Economic Geography in Oxford. The authors are grateful to participants at these sessions, as well as to the editor and two anonymous referees for Industry & Innovation, for advice that helped to improve the paper.

Notes

¹ This has sometimes been referred to as relational proximity (Coenen, Moodysson, and Asheim 2004) or personal proximity (Schamp, Rentmeister, and Lo 2004).

² The assessment of the most important partner is based on the respondent’s subjective view of which partner has contributed the most to new product or process development (i.e. innovation). We do not have detailed information on how many partners the firm has. However, the telephone survey asked whether firms collaborated with partners of seven different types (internal, suppliers, customers, competitors, consultancies, universities and research institutes) and at three different scales (regional, national and international). From the 21 different types of partner this potentially yields, the average firm collaborated with 5.4 types, with a standard deviation of 3.3. The first quartile was 3 types, the median 5 and the third-quartile 7 types of partner. Consequently, most firms select one partner perceived as the most important from a broader portfolio of several partners.

³ ‘Elsewhere in Norway’ was selected as medium proximity as it seemed useful to differentiate between national and international collaborations from a geographic perspective. As a robustness check, we have also estimated all models treating ‘in the same municipality’ as high proximity, ‘in the same region’ as medium proximity, and outside the region as low proximity. These analyses show very similar results for high and medium proximity, suggesting that it is reasonable to treat local and regional partners as a single category. In a separate analysis, we estimated the models treating Norwegian partners as high proximity, other Scandinavian partners as medium proximity and non-Scandinavian partners as low proximity. The results are similar to those reported in the paper. The levels of innovation were slightly higher for Scandinavian than (extra-regional) national partners, but firms using non-Scandinavian partners reported the highest levels of innovation. We have reported the main differences in the findings using these alternative specifications at the relevant points in the paper.

⁴ The three lower categories on the Likert scale are not very prevalent in the data and we therefore combined them into one category. The overall rationale of our categorisation was that the low proximity category is the below median category, high proximity is above median, while medium proximity is the median category. As a robustness check, we have also respecified the analysis treating ‘neither agree nor disagree’ as medium proximity. In these analyses, medium proximity is not associated with innovation, suggesting that firms in this category are still too far from their partners and that the ‘optimal’ distance is near the ‘partly agree’ category.

⁵ As a robustness check, we have estimated all models for the subsample of 410 firms that responded to all the proximity questions. These analyses show very similar results to those reported in the paper. They can be made available upon request.

⁶ In one of the robustness checks, we ran the analyses using ‘elsewhere in Scandinavia’ as the medium category, as reported in note 3. These analyses show a significant association between geographic proximity and social and institutional proximity (at the 95 per cent level), as well as cognitive proximity (at the 90 per cent level).

⁷ The results are similar when Scandinavia is treated as medium proximity, while they provide even stronger support for the compensation mechanism hypothesis when ‘in the same region’ is treated as medium proximity. In the latter case, all combinations of high non-geographical and low geographical proximity have a significant effect, as do three combinations of medium non-geographical and low geographical proximity. Conversely, no combinations of low non-geographical and low geographical proximity are significantly associated with innovation in this specification.